Development and optimization of the low-cost optical monitor for real-time monitoring of atmospheric black carbon

• Machine learning models trained by using spectrum values for Black retrieval. • The GBR model performed better than the RF model in estimating BC, yielding more accurate results. • Machine learning models demonstrate high accuracy in predicting BC. • Carbon Scan is a cost-effective method to monitor BC on large temporal and spatial regions. • Carbon Scan overcomes low-cost methods by enabling continuous and real-time monitoring. Black Carbon (BC) is a major component of atmospheric aerosol produced from incomplete combustion of fossil fuels, biomass, and solid fuel. About 42 % of BC emissions are from open biomass burning with major contributions from Africa and Asia. A low-cost (<1500 US$) monitoring system is designed for large scale monitoring atmospheric BC near the surface for effective mitigation actions. Carbon Scan includes an air sampler with filtration of particulate matter of size 2.5 micron or less, a color sensor that captures image, and a machine learning (ML) model that retrieves BC concentration. Gradient Boosting Regressor (GBR) and Neural Network (NN) were trained and evaluated for the retrieval of atmospheric BC and its fraction from biomass burning. Carbon scan is a significant advancement in atmospheric BC monitoring. It achieves great accuracy, with a low RMSE of 1 µg/m³, Mean Absolute Error (MAE) of 0.40 µg/m³, Mean Absolute Percentage Error (MAPE) of 6.76 %, (SMAPE) of 5.84 % and a high R² of 0.97. The sensor provides an opportunity to monitor real time concentrations of atmospheric BC. Carbon scan is a low power, low cost, that ensures continuous air monitoring in remote areas, while capturing large temporal and spatial variations of BC.